Pairs‐trading and spread persistence in the European stock market

• Author: Tao Tang,Isabel Figuerola‐Ferretti,Ioannis Paraskevopoulos
• DOI: http://doi.org/10.1002/fut.21927
• Date: 01 September 2018
• Published date: 01 September 2018

Received: 1 October 2016

|

Revised: 21 March 2018

|

Accepted: 1 April 2018

DOI: 10.1002/fut.21927

RESEARCH ARTICLE

Pairs‐trading and spread persistence in the European

stock market

Isabel Figuerola‐Ferretti

1

|

Ioannis Paraskevopoulos

2

|

Tao Tang

3

1

Quantitative Finance Group, Finance

Department, ICADE Universidad

Pontificia Comillas, Madrid, Spain

2

Capital Markets Department, Bankia

Bank, and Quantitative Finance Group

ICADE, Universidad Pontificia Comillas,

Madrid, Spain

3

Finance Department and Institute of

Finance, College of Economics, Jinan

University, Guangzhou, China

Correspondence

Tao Tang, Finance Department and

Institute of Finance, College of

Economics, Jinan University,

Guangzhou 510632, China.

Email: taotang@jnu.edu.cn

Funding information

Spanish Ministry of Economy, Industry

and Competitiveness, MINECO,

Grant/Award Numbers: ECO2016‐78652,

ECO2013‐46395

In this paper, we adapt the demand and supply framework introduced by

Figuerola‐Ferretti and Gonzalo (Journal of Econometrics, 2010) to illustrate the

dynamics of pairs‐trading. We underline the process by which a finite elasticity

of demand for spread trading determines the speed of mean reversion and pairs‐

trading profitability. A persistence‐dependent trading trigger is introduced

accordingly. Applied to STOXX Europe 600–traded equities, our strategy

exploits price leadership for portfolio replication purposes and delivers Sharpe

ratios that outperform the benchmark rules used in the literature. Portfolio

performance and mean reversion are enhanced after firm fundamental factor

restrictions are imposed.

KEYWORDS

cointegration, error persistence, pairs‐trading, price discovery, trading trigger

JEL CLASSIFICATION

C58, G11, G12, G14

1

|

INTRODUCTION

Short‐term price discrepancies are common across assets that are imperfectly integrated. Pairs‐trading strategies are

designed to earn profits from relative mispricings of closely related assets. This paper exploits commonalities arising

from cointegrated assets to model relative value arbitrage via pairs‐trading strategies. Pairs‐trading belongs to the family

of convergence trade strategies. It relies on a well‐known trading rule for cointegrated price series based on

simultaneous long–short positions that are closed when prices revert to a long‐run relationship. When an investor

opens a position he shorts the outperformer and longs the underperformer, until the mispricing is eliminated. We

extend the Figuerola‐Ferretti and Gonzalo (2010) (FFG hereafter) demand and supply framework to describe price

dynamics in two distinct but cointegrated assets and show how market participants exploit temporary mispricings

performing pairs‐trading strategies. The setup requires a finite elasticity of arbitrage services and cointegration error

persistence. It evolves around the speed by which arbitrageurs restore equilibrium allowing the measurement of price

discovery for portfolio replication purposes and arbitrage profit determination. A market is regarded as dominant in this

framework if it concentrates a larger number of participants. Cointegration, therefore, guarantees price convergence

that is represented in terms of a stationary error correction term. A trading trigger is derived, which is linked to the

degree of persistence of the cointegration error so that a higher stationarity requires a lower trading trigger.

This paper is relatedto Gatev, Goetzmann, and Rouwenhorst(2006) (GGR hereafter), who examinethe performance of

pairs‐trading using the daily U.S. stock return data. GGR performs pairs selection using the minimum‐distance algorithm.

J Futures Markets. 2018;38:998–1023.wileyonlinelibrary.com/journal/fut998

|

© 2018 Wiley Periodicals, Inc.

They find economically and statistically significant excess returns of around 11% per annum. Following GGR, Andrade, di

Pietro, and Seasholes (2005); Broussard and Vaihekoski (2012); and Bowen and Hutchinson (2016) apply the algorithm to

Asian and European equity markets. A common drawback from these studies is that they essentially apply an ad hoc

trading trigger. Vidyamurthy (2004) sheds light on this problem by searching for a trading trigger optimality by

maximizing a profit function under the assumption of Gaussian errors.

Another strand of the literature models the cointegration spread under dynamic settings. Elliott, van der Hoek, and

Malcolm (2005) and Avellaneda and Lee (2010) consider an Ornstein–Uhlenbeck process to model the cointegration error

allowing spread estimation and setting the framework for the determination of the threshold value. While Avellaneda and

Lee (2010) empirically determine cutoff values based on the process assumed for the cointegration error, Elliott et al. (2005)

link the trading trigger to the degree of mean reversion. This paper contributes to the literature by adapting the FFG model

to paired equity prices to illustrate the process of cointegration error correction by means of an economically meaningful

vector error correction model (VECM). In doing this, we show that pairs‐trading profitability is dependent on the speed of

adjustment (or spread mean reversion), which is determined by the elasticity of demand for pairs‐trading strategies and the

total number of market participants. Therefore, we demonstrate that lower error persistence leads to higher pairs‐trading

profitability. We accordingly propose a trading trigger that is determined by the speed of convergence to the

long‐run stationary relationship arising from the VECM estimates. Our model‐based trading rule is, therefore, related to

Elliott et al. (2005) in that the trading trigger is defined as a function of the speed of mean reversion. This is motivated from

VECM dynamics as pairs‐trading profitability is directly dependent on cointegration error persistence, which is determined

by the speed of mean reversion. This is consistent with the results reported by Kanamura, Rachev, and

Fabozzi (2010), in which an empirical link between profitability and mean reversion is established for spread trading in

the gas market. The relationship between the speed of adjustment and the number of participants has been empirically

addressed by Brennan, Jegadeesh, and Swaminathan (1993) in a study relating the number of informed traders, proxied by

the number of analyst following a firm, with the speed of adjustment to common shocks.

Our empirical application is based on an out‐of‐sample analysis and uses STOXX Europe 600–traded equities whose

prices are quoted in the euro currency to identify cointegration relationships with a sample ranging from 2000 to 2017.

Common factor and industry restrictions are imposed to illustrate the existence of long‐run stationary relationships.

This justifies the use of a model with equity shared fundamentals that drive prices to parity. We use price leadership for

portfolio replication purposes in an extensive out‐of‐sample estimation. We analyze the profitability of pairs strategies at

the portfolio level and compare their performance with benchmark pairs‐trading methodologies used in the literature.

We find that the proposed pairs strategies outperform the seminal strategy of Gatev et al. (2006), as evidenced by

significant abnormal returns and higher Sharpe ratios. The documented outperformance is enhanced once we control

for common firm fundamentals as well as the industry effect.

The rest of the paper proceeds as follows: Section 2 relates the VECM dynamics to the construction of pairs‐trading

strategies, which requires a description of preliminaries and main results of the FFG model applied to two distinct but

cointegrated assets; Section 3 presents the data sample and empirical results on cointegration and price discovery;

Section 4 shows the pairs‐trading performance analysis with a number of robustness tests developed to illustrate the

outperformance of our model‐based approach; Section 5 provides conclusions.

2

|

THE THEORETICAL MODEL

2.1

|

Model setup

The aim of this section is to introduce pairs‐trading strategies in a demand and supply framework. Paired firms in this

context share common fundamentals and are linked via long‐run stationary relationships generated by market forces.

Accordingly, arbitrage takes place through pairs‐trading strategies that exploit mean reversion of pricing errors.

Convergence takes place because paired assets measure a common underlying factor. The model is built on the

presumption that price correction of two cointegrated assets departing from long‐term stationary relationships depends

on the average speed of convergence in each market. This determines the degree of persistence of the cointegration

error and becomes an important factor for both the trading trigger and profit determination.

In what follows, we present the joint dynamics between two cointegrated assets within a demand and supply market

clearing framework. Investors either take single asset positions or trade two assets that share common fundamentals

simultaneously via the use of pairs‐trading strategies. The mean reversion of the cointegration spread is of critical

importance to arbitrageurs who will exploit short‐lived deviations in search of profitability from pairs strategies.

FIGUEROLA‐FERRETTI ET AL.

|

999